A PRELIMINARY EXPERIMENTAL STUDY OF CNN-BASED BRAIN TUMOR CLASSIFICATION AND U-NET-BASED SEGMENTATION FROM MRI
: https://doi.org/10.65725/JCISE/2/3/013
JOURNAL OF COMPUTATIONAL INTELLIGENCE SCIENCE AND ENGINEERING (JCISE) 
ISSN: 3107-8168
Volume 2 Issue 3, Jul-Sep 2026

Abstract:
Accurate analysis of brain tumor regions from magnetic resonance imaging (MRI) is an important task in automated medical image analysis. Deep learning methods provide effective approaches for tumor classification and pixel-level segmentation of MRI images. This study presents a preliminary experimental investigation combining a convolutional neural network (CNN) for brain tumor classification with U-Net-based architectures for 2D brain tumor segmentation. The CNN component performs image-level tumor classification, while U-Net and Attention U-Net are employed for tumor-region segmentation. The reported CNN classification results achieved an accuracy of 94.2%, with precision and recall values of 0.91 and 0.88, respectively. For the segmentation models, the reported U-Net results achieved an accuracy of 96.8%, Dice coefficient of 0.92, precision of 0.94, and recall of 0.91, while Attention U-Net achieved an accuracy of 98.5%, Dice coefficient of 0.95, precision of 0.97, and recall of 0.94. These results provide a preliminary experimental assessment of CNN-based classification and U-Net-based segmentation approaches for brain tumor MRI analysis. The study is intentionally presented as a basic experimental investigation rather than a novel architectural contribution and provides a foundation for subsequent research involving advanced architectures and volumetric 3D brain tumor segmentation..

Authors: Felicia Rajathi P and Dr.D.Gladis

Keywords: Brain Tumor, MRI, Brain Tumor Classification, Brain Tumor Segmentation, CNN, U-Net, Attention U-Net, Deep Learning, Medical Image Analysis.